Optical lithography below 100nm
11/01/1998
SECOND IN A SERIES
Optical lithography below 100 nm
John H. Bruning, Tropel Corp., Fairport, New York
Optical lithography has been the primary patterning technology for IC production for 30 years. Momentum and astounding improvements in resists and optics have forestalled the introduction of more complex alternatives. How much longer is this likely to continue? Various improvement paths with alternate lens and system tradeoffs indicate that there is still headroom left, and that the end of optical lithography is below 100 nm - probably close to 70-80 nm.
Several alternatives exist for leveraging the remaining headroom in optical lithography. Numerical apertures (NAs) can be raised still closer to unity and wavelength can be reduced further to 157 nm, still using known optical designs and bulk refractive optical materials. Resolution enhancement techniques, regardless of the wavelength or NA, require the ultimate in performance from the imaging lens. Sub-100-nm optical lithography could result via several alternate paths, but the particular approach chosen for production will depend specifically on the development rate and progress of resist technology, photomasks, optical technology, and scanning strategies and technology.
Examining some of the recent optical lithography system parameters (Table 1), we see that dramatic changes have taken place, particularly to the k1 factor used in production [1]. This parameter, which essentially indicates a process capability associated with the resist, the mask, and the exposure tool, has shown dramatic improvement up to the present and is likely to show further improvement with a limit probably somewhat below 0.4. The ability to raise NA much above 0.6 requires re-examining some exposure system assumptions, in particular, how to trade off field size of the lens with the exposure technique of the lithography tool.
It is clear from the data in Table 1 that the sensitivity of the entire lithographic process is very high and increasing. The k1 factor and the depth of focus (DOF) are very small. As the critical dimension (CD) requirements become more aggressive, CD uniformity over the die must also improve in proportion to the CD itself. Lens imperfections from design and manufacturing must all improve to an extent that is much more important now, and in the future, than it was in the past. This improvement becomes increasingly difficult, if not contradictory, since greater performance must usually be accompanied by greater complexity, tighter manufacturing tolerances, better materials, and higher cost.
 |
What limits lens performance?
It is instructive to examine what limits ultimate lens performance and how lithographic system parameters can be optimized to achieve it. One of the greatest levers to lens performance is scale. A microscope objective, for example, can be made to deliver diffraction-limited performance over a broad wavelength range at NAs as high as 0.95. Microscope objectives, however, have very small fields, usually less than 1 mm.
Microlithographic lenses, on the other hand, are considerably more demanding because they have much larger fields along with very tight requirements on flatness of the image field, uniformity, and distortion. In addition, the image at the wafer must be telecentric, meaning that the cone of rays from each object point on the reticle must land nominally perpendicular to the wafer so that waviness of the wafer surface does not translate into lateral misalignments of the wafer image. Field flatness, telecentricity, and distortion generally drive the design form and complexity of lithographic lenses.
Lithographic lenses must be diffraction limited, meaning geometric aberrations determined by tracing rays through the many air-glass interfaces must be dimensionally less than the wavelength of exposure. The stringent demands on today`s ICs for precise CD control and image placements go well beyond the simple specification of "quarter-wave wavefronts." Generally, the quality of the wavefront at any point in the image field must be <0.04 wave (rms). Different aberrations have different effects on the image. "Sweet-spots" in the image give rise to CD variations that are unacceptable. Resolution extension techniques will drive wavefront requirements down to the 0.02-0.03 wave (rms) level if they are to work predictably and reliably. These very aggressive specifications are seldom if ever reached in practice, but will be required if optical lithography is to reach its full potential.
Wavefront aberrations in a lens design on paper are determined by ray tracing and have characteristics that are a result of the particular optical design "form" - the arrangement and shapes of the many lens elements that make up the complete lithographic lens. Wavefront aberrations of a manufactured lens are a composite of the residual aberrations from the paper design, and the quality of the optical materials, manufactured components, antireflection coatings, and lens assembly process. All these components contribute to wavefront quality and are proportional to the scale of the lens, which in turn is proportional to the field size of the lens. Most critical lens characteristics vary directly with scale or field size (F), as listed in Table 2. This table gives strong clues to the burden carried by the large field requirement of contemporary lithographic lenses.
 |
If the lens were built at half-scale, theoretically, all geometrical aberrations would be half their value. In practice, the situation is somewhat better, as bulk material absorbance would also be half, meaning lens transmission would be higher. Similarly, inhomogeneity of the material would be half since the optical path through the material is half. The cost of the material would be roughly 1/8, since the volume or mass of material used is proportional to the cube of the scale or field. The composite wavefront aberration should improve more than a factor of two because of improved material homogeneity. In reality, a half-scale lens would not be scaled quite so simply: the designer would optimize the balance of residual aberrations for the particular situation and would probably reduce the number of lens elements as well. Reducing the number of lens elements also has a positive effect, as there are fewer surface contributions, each with their own residual surface errors, surface scatter, and homogeneity errors.
Making higher-resolution lenses
To extend the capability of optical lithography, we need to improve the wavefront quality and resolution of lenses without having to make smaller chips. In other words, we need to give up field-size without giving up die-size. The way around this dilemma is simply to reduce the scale or field-size of the lens. In fact, just such a transition occurred in the early 1980s, when optical-lithography systems moved from imaging the entire wafer without reduction (i.e., 1:1 using 1? reticles) to imaging a small cluster of chips at either 1:1 or 10:1 reduction [2].
Around 1996, the introduction of step-and-scan systems [3-4] marked another transition to smaller field systems by replacing the classical stepper lens with a lens that would image a narrow vertical strip spanning the field diameter of the lens. As shown in Fig. 1, a stepper lens with field-diameter D, used in a step-and-scan mode, can expose a die height approximately 30% larger than when used in a conventional stepper mode. Conversely, a lens with a 30% smaller field can expose, by step-and-scan or 1D scanning, a die of the same size as the field of a larger stepper lens.
Scanning improves uniformity
Scanning has the added benefits of improving DOF and CD uniformity [5-6]. Waviness on a wafer in the scan direction is tracked with an autofocus system rather than being set to a single average value for the entire stepper lens. Better focus control, by itself, results in better CD control. In the scanning direction, CD uniformity is improved because the same physical portion of the lens field is used to expose all horizontal positions of the die. There is no field variation in the scanning direction; image quality is an average over the width of the scanning strip.
Additional improvements in lens performance can be made by partitioning the die height into several overlapping strips (Fig. 2). If the die field is divided vertically into two or three horizontal strips, the field size or scale of the lens can be reduced to a half or third of that used in a step-and-scan lens. This reduction allows additional performance and capability to be built into the lens without adding complexity (i.e., more lens elements) and cost.
 |
Figure 1. Comparison of lens field-size vs. die-size for a) steppers and b) scanners.
 |
Figure 2. A comparison of exposure fields and methods for a) step-and-scan and b) raster-scan.
A simple optical principle helps illustrate the point: The complexity of an optical system may also be described by its optical invariant, which is the product of the maximum image field height and the maximum ray angle at the image. The optical invariant, by definition, is the same on the object or reticle side of the lens as on the image or wafer side. This same principle in essence says that if we make small changes to the lens, e.g., reduce the field size by x%, the NA can be increased by the same x%. The product of the new field size and the new NA remains unchanged or invariant.
If we reduce field size by a factor of two, the implication is that we could double the NA. This is true, however, only for lenses of very small NA. For high-resolution, high-NA lithographic lenses, the situation is not quite so simple, but the general conclusion is valid: reducing the field is a powerful approach to increasing the NA and performance of the lens, provided the required field coverage can be accomplished in some other way (such as scanning).
As mentioned previously, this approach was first used when wafers became too large to expose in a single lens field, and once again with the move to step-and-scan exposure to accommodate chips now too large to fit within a single stepper lens field. Further partitioning of the image field at this point is not required to make larger die, but to unburden the lens of field size so that it can operate at higher NA, greater performance, lower cost, or some combination of these attributes.
The power of a smaller field lens
There was an industry need several years ago for an advanced lens and cost-effective lithographic tool for the development of processes and resists to prepare for the eventual introduction of production 193-nm lithography. The requirements of the lens were for good imagery with variable NA up to 0.6, along with the flexibility to explore resolution extension techniques and illumination alternatives. Field size was not a key concern, since this tool was not intended for device fabrication. The lens was designed for ultimate imaging performance so resolution extension techniques could be exploited independent of artifacts. The lens was also optimized to work over a relatively broad bandwidth of 0.5 nm so that the ArF excimer laser source could be used without bandwidth narrowing. The lens designed for this purpose, with its relatively small 2.1-mm field, permits a unique optical design form that virtually guarantees that the performance on axis is identical to its performance at the edge of the field [7]. The lens has only five elements compared to the more usual number of 20-30 in a state-of-the-art lithographic lens. The design performance of this lens
 |
Figure 3. Performance attainable when a lens is optimized for image quality. The images were made in a resist developed by the University of Texas using a Levenson phase-shift mask fabricated at DuPont. Images show
a) 80-nm lines and 120-nm spaces, b) 90-nm lines and 135-nm spaces [8].
is approximately 0.01 wave, and the measured as-built performance is 0.02 wave. This performance level is unattainable in a large-field lithographic lens.
The demonstrated results with this simple lens, optimized for imaging performance, have been quite spectacular [8]. Figure 3 shows results from one of these systems at SEMATECH using a newly developed resist from the University of Texas. The role of the resist and the photomask cannot be overstated in these results and others, particularly as we approach the limits of optical lithography. The resist images show a k1 factor as small as 0.25 (Fig. 3a). This is both a triumph of resist technology and phase-shift mask technology. These results may indeed be difficult to duplicate in a production environment.
Beyond step-and-scan
Conventional step-and-scan systems, most of which evolved from steppers, did so by adding a 1D x-scanning stage at the reticle. This increased overall system complexity, but required a lens with a 30% smaller field.
Because of scanning, there are additional overhead times that reduce throughput. With fast stages, throughput is ultimately limited by resist sensitivity and exposure intensity at the wafer. If an additional y-motion is added to the reticle stage, a large vertical field can be exposed in overlapping strips, as previously illustrated in Fig. 2b. This exposure technique provides complete independence between lens field-size and die-size. This small additional system complexity frees up enormous potential for higher-resolution lenses and adds further longevity to optical lithography. The use of a small field lens allows a flatter field, lower distortion, and higher performance, while 2D scanning of a smaller field allows more precise local focus control and greater CD uniformity. Because overlay is also determined by laser metrology-based scanning stages, tool matching becomes less dependent on individual lens distortion signatures [5].
Two-dimensional scanning is not a fundamentally new concept [9-10]. For example, lithographic imaging has been demonstrated using a microscope objective at 10:1 reduction to cover a 1 ? 1-cm die area by raster-scanning the reticle and wafer with overlapping exposures [9]. To extend this concept to practical high-throughput lithography requires optimizing the shape and size of the overlapped image field. Figure 4 shows throughput calculations for 300-mm wafers with two different exposure field configurations: 4a shows a throughput calculation with a narrow hexagonal-shaped field; 4b shows the same calculation with a regular hexagonal exposure region. The overhead times are larger in 4b than in 4a, even though the field of the lens is used more efficiently in 4b. Some other factors affecting the choice of the exposure field shape are the maximum velocity and acceleration of the stages, and the minimum allowable number of laser pulses/field point (to achieve the required uniformity of exposure). Smaller exposure fields result in more accurate focus control and better CD control, but reduce throughput. Fully considered choices should be driven by a thorough cost-of-ownership model.
Is there a role for 157 nm?
Perhaps the greatest risk for 157-nm lithography is the need for mask technology to change from a robust and entrenched fused-silica technology to a new and untried MgF2 or CaF2-based technology. The overall burden to the infrastructure may be greater than the benefit of the wavelength reduction from 193 nm to 157 nm. Mask issues aside, in this author`s opinion, 157-nm lithography lenses may not be practical at the scale of today`s lenses, but at reduced scale are certainly manufacturable. Antireflection coatings at 157 nm are not at the same mature state of development as they are for 248 nm and 193 nm because there are no known high-index materials that transmit well at 157 nm.
A promising new design form applicable to the raster-scan concept discussed here is shown in Fig. 5. This lens has an NA of 0.75 and an image field diameter of 8 mm. The simple construction is similar to that previously described [7], but with a 4:1 reduction ratio. The lens is color-corrected with a single material by incorporating a coaxial beam-splitter. This same design is nearly identical for use at both 193 nm and 157 nm simply with a material change. By coincidence, the refractive index of fused silica at 193 nm is nearly identical to the refractive index of CaF2 at 157 nm. The bulk transmission of high-grade fused silica at 193 nm and CaF2 at 157 nm are also nearly identical. This design form is a powerful candidate for a color-corrected solution at either of these wavelengths in a modest field size configuration and would be considerably less costly to build than contemporary lenses. The decision as to whether 157-nm lenses are built and deployed will probably be delayed for some time to come.
Where does EUV fit?
Extreme ultraviolet (EUV) is the most sensible lithographic approach to follow when optical lithography requires all reflecting surfaces in the imaging lens. Since wavefront quality of the
 |
Figure 4. Throughput of 300-mm wafers by raster-scanning: a) the exposure field is a narrow hexagonal strip with a small overlapped scan area; and b) the overlapped area is considerably larger; the full circular field of the lens is used efficiently, but there is considerably more scanning overhead time. Laser power is 3 W; resist sensitivity is 10 mJ/cm2; illumination or transmission efficiency is 12%, maximum acceleration is 1 g, and maximum reticle stage velocity is 1 m/sec.
 |
Figure 5. A new design form for 193-nm or 157-nm raster-scan lithography. This lens is color-corrected and has only six elements (the largest, 200 mm dia.).
lens, transmission through the lens, and brightness of the source represent significant challenges to this technology, a little thought suggests that reducing the scale of the lens is an appropriate and powerful approach to reducing cost and improving performance for EUV as well. Reducing the scale reduces all design aberrations (all imaging aberrations, field flatness, and distortion) by the scale factor, making the lens easier to fabricate for a variety of reasons. First, the aspheric departure of the mirror surfaces is reduced by the scale factor. Next, the surfaces themselves are reduced in size by the scale factor. Since the design aberrations of a reduced scale system are smaller, the NA of the system could be increased without adding additional surfaces. A physically smaller lens system would be mechanically more stable, more manufacturable and less expensive. Lessons learned by raster scanning in the deep-UV would all be applicable in the EUV.
Conclusion
If the full potential of optical lithography is to be exploited, the design of current lithographic systems needs to include 2D or raster-scanning with smaller field lenses. This small evolutionary step will permit the manufacturing of virtually perfect lenses of very high NA with a minor penalty to throughput. This is the only way that lenses can be made good enough to exploit fully the potential of resolution-extension techniques. Future challenges may include polarization artifacts at very high NAs, the need for top surface imaging resists, and line-edge roughness.
Raster scanning should increase the longevity of optical lithography and provide a cost-reduction path and more performance headroom. All these issues must be explored more fully with detailed cost-of-ownership models.
With so many aspects of the lithography process improving and under investigation, it remains difficult to predict with any certainty which particular path will lead us to sub-100-nm optical lithography. Perhaps most significant, we need to keep in mind that we have never yet overestimated the future capability of improvements in optical lithography. n
References
1. T. Brunner, "Pushing the Limits of Lithography for IC Production," Proc. IEDM 97, pp. 9-13, 1997.
2. J. Bruning, "Optical Lithography - Thirty Years and Three Orders of Magnitude," SPIE, Proc. Vol. 3048, pp. 14-27, 1997.
3. D. Cote, et al., "Micrascan III - Performance of a Third Generation, Catadioptric Step and Scan Lithographic Tool," SPIE, Proc. Vol. 3051, p. 806, 1997.
4. G. de Zwart, et al., "Performance of a Step and Scan System for DUV Lithography," SPIE, Proc. Vol. 3051, p. 817, 1997.
5. M. van den Brink, et al., "Step-and-Scan and Step-and-Repeat, a Technology Comparison," SPIE, Proc. Vol. 2726, p. 734, 1996.
6. W. Arnold, "Is a Scanner Better Than a Stepper?" Solid State Technology, p. 77, March 1997.
7. J. Webb, J. Nemechek, "Optical fabrication rises to the 193-nm challenge," Laser Focus World, Feb. 1997.
8. J. Byers, et al., "Recent Advancements in Cycloolefin Based Resists for ArF Lithography," J. Photopolymer Sci. and Tech. 11(3), pp. 465-474, 1998.
9. G. Bulnov, N. Kontievskaya, "Evaluation of Effect of Scanning on the Resolution of Optical Systems," Sov, J. Opt. Technol., 48(9), pp. 566-567, 1981.
10. K. Jain, "A Novel High-Resolution, Large-Field Scan-and-Repeat Projection Lithography System," Proc. SPIE, Vol. 1436, pp. 666-677, 1991.
JOHN BRUNING received his BS from Penn State University and his MS and PhD from the University of Illinois, all in electrical engineering. He has published widely in optical lithography and optical metrology and has more than 20 patents. Bruning is president and CEO of Tropel Corp., 60 O`Connor Rd., Fairport, NY 14450; ph 716/388-3400,
fax 716/377-6332, e-mail [email protected].